A method for predicting the association between microorganisms and diseases based on complementary fusion of dual association graphs

Abstract

This invention belongs to the field of biomedical technology, specifically a method for predicting the association between microorganisms and diseases based on complementary fusion of dual association graphs. The specific steps include: acquiring association information between microorganisms and diseases, constructing a first microorganism-disease heterogeneous association graph, represented as matrix B1; acquiring association information between microorganisms and drugs, and drugs and diseases, constructing a second microorganism-disease heterogeneous association graph, represented as matrix B2. This invention effectively integrates multi-dimensional feature information through a graph-enhanced feature fusion extraction module, improving the accuracy and comprehensiveness of feature representation. Furthermore, the complementary fusion model, based on the microorganism-disease dual association graph fusion matrix, combines a complementary attention mechanism and a soft-label KL divergence loss function to effectively solve the problem of insufficient fusion of information from different association graphs, overcoming the limitation of limited complementarity between association graphs.

Description

Technical Field

[0001] This invention relates to the field of biomedical technology, specifically to a method for predicting the association between microorganisms and diseases based on complementary fusion of dual association graphs. Background Technology

[0002] In recent years, predicting the association between microbes and diseases has become a research hotspot in the biomedical field. Changes in the microbial community are considered to be closely related to the occurrence of various diseases, such as the relationship between the gut microbiota and obesity, diabetes, and inflammatory bowel disease. Therefore, how to effectively reveal the association between microbes and diseases has become an important topic in disease research and precision medicine.

[0003] To improve the accuracy, interpretability, and scalability of predicting associations between microorganisms and diseases, graph structure models have gradually become an effective tool, widely used in research on the relationship between microorganisms and diseases. Feature extraction based on graph structures is a crucial step in predicting associations between microorganisms and diseases. Current techniques primarily rely on matrix factorization and Graph Convolutional Networks (GCNs) for feature extraction, but both have limitations. First, matrix factorization methods tend to overlook the global structure and local details of the data during dimensionality reduction, leading to a significant loss of original information and insufficient representation of the data. Second, while GCNs can extract high-order features through message passing mechanisms in graph structures, as the number of layers increases, the features of nodes tend to become similar, resulting in insufficient representation of the differences between nodes. Furthermore, the feature aggregation methods of GCNs often fail to fully integrate multi-dimensional features.

[0004] As research deepens, association graph fusion technology has gradually emerged as a new research direction. Association graph fusion technology improves the predictive accuracy of the association between microorganisms and diseases by fusing information from different association graphs. Currently, association graph fusion methods mainly utilize deep learning frameworks to integrate and propagate information in the modeling of graph-structured data. In this way, the model can capture the complex relationships between nodes in multiple association graphs, achieving information propagation and feature learning. However, these methods still have the following drawbacks:

[0005] 1. Existing technologies for predicting the association between microorganisms and diseases suffer from incomplete integration of multi-dimensional features during the feature extraction stage, resulting in an incomplete feature set that fails to fully reflect the complex intrinsic relationship between microorganisms and diseases;

[0006] 2. In the process of obtaining topological attributes, the existing restart random walk method has the problems of insufficient information propagation and incomplete capture of long-distance relationships;

[0007] 3. In existing association graph fusion methods for predicting the association between microorganisms and diseases, simple splicing or weighted summation is usually used, lacking an effective mechanism for information interaction and fusion of association graphs. This results in the interaction between different association graphs not being fully optimized, and the information from different association graphs cannot be effectively fused, ultimately affecting the overall performance and prediction accuracy of the model.

[0008] 4. Existing association graph fusion methods for predicting the association between microorganisms and diseases often fail to fully explore the complementary information between the association graphs while maintaining the uniqueness of each graph. This makes it difficult to effectively distinguish between redundant and unique information during the fusion process, leading to information conflicts and the propagation of redundancy, which ultimately affects the overall performance of the model and reduces the accuracy and robustness of the prediction.

[0009] Therefore, a method for predicting the association between microorganisms and diseases based on complementary fusion of dual association graphs is invented. Summary of the Invention

[0010] To address the aforementioned technical problems, according to one aspect of the present invention, the present invention provides the following technical solution:

[0011] A method for predicting the association between microorganisms and diseases based on complementary fusion of dual association graphs includes the following specific steps:

[0012] S1: Obtain the association information between microorganisms and diseases, construct the first microorganism-disease heterogeneous association graph, represented as matrix B1; obtain the association information between microorganisms and drugs, and drugs and diseases, construct the second microorganism-disease heterogeneous association graph, represented as matrix B2;

[0013] S2: Apply the graph-enhanced feature fusion extraction module GEFFE to obtain the feature matrix Z1 corresponding to B1 and the feature matrix Z2 corresponding to B2; the specific steps are as follows:

[0014] S21: Based on protein function associations, calculate for any given m i and m j Functional similarity between them, constructing a microbial functional similarity matrix. Based on gene interaction information, the calculation is performed for any given d. i and d j Functional similarity between them, constructing a disease functional similarity matrix Where nm and nd represent the number of microorganisms and diseases, respectively;

[0015] S22: Perform a topology-enhancing restart random walk operation TERRW on matrices ZM1, ZM2, ZD1, and ZD2 respectively to obtain the first microbial topology attribute matrix. Second microbial topological attribute matrix First Disease Topological Attribute Matrix Second disease topological attribute matrix The TERRW operation is defined as follows:

[0016]

[0017] in, c represents the walk probability distribution of the i-th node in the l-th step; i It is the dynamic adjustment probability calculated by node i based on its local topology information; It is the extended transition probability matrix; ∈ i It is the initial probability vector of node i;

[0018] S23: To preserve more original features, the functional similarity, topological attributes, and adjacency relationships between microorganisms and diseases are concatenated to construct the feature matrix corresponding to B1. The characteristic matrix corresponding to B2 Its Z1 and Z2 are defined as follows:

[0019]

[0020]

[0021] Where A1 represents the microorganism-disease adjacency matrix A1, and A2 represents the microorganism-disease adjacency matrix A2;

[0022] S3: Using B1 and B2, construct the dual-association graph fusion matrix B. Simultaneously, based on B, Z1, and Z2, apply the dual-association graph complementary fusion model DAG-CFM to update node information, obtaining the node collaborative representation matrix H corresponding to B. The specific steps are as follows:

[0023] S31: Construct a dual-association graph fusion matrix B based on B1 and B2;

[0024] S32: Based on B, apply the dual association graph complementary attention mechanism DAG-CA to calculate the attention weight between each node and its neighboring nodes in the same association graph and the attention weight between its neighboring nodes across association graphs.

[0025] S33: Combine the neighbor node weights obtained from Z1, Z2 and DAG-CA to perform bi-association graph information fusion and construct the node collaborative representation matrix H corresponding to B;

[0026] S34: The model uses the Adam optimizer, minimizing the soft-label KL divergence loss L. KL To train collaborative representation learning; soft-label KL divergence loss L KL It is used to measure the difference between the target distribution and the soft label assignment distribution, and is defined as follows:

[0027]

[0028] Where C is the number of clusters; t eo This is the soft label assignment probability, representing the probability that node e belongs to cluster o; q eo It is the target distribution, obtained by squared normalization of the soft label assignment probability;

[0029] t eo The calculation process is as follows:

[0030]

[0031] Where, μ o The center of cluster o is randomly initialized, and for each cluster center it is initialized as a random node in H; ||*|| represents the Euclidean distance;

[0032] q eo The calculation process is as follows:

[0033]

[0034] S4: Using H, the specific steps for predicting the association between microorganisms and diseases through MLP are as follows:

[0035] S41: Using H, define the collaborative representation vector H for each microorganism. i Collaborative representation vector H with each disease j The correlation between them to MDA ij MDA ij The calculation is as follows:

[0036] MDA ij =-||H i -H j ||

[0037] Among them, H i -H j This indicates the difference between a microorganism and a disease; the smaller the difference, the higher the likelihood that the microorganism-disease pair is associated.

[0038] S42: Utilizing MDA ij Using MLP to determine the predictive association probability of microorganisms-diseases ij The specific methods are as follows:

[0039] ass ij =MLP(MDA) ij ).

[0040] As a preferred embodiment of the microbial-disease association prediction method based on complementary fusion of dual association graphs described in this invention, the specific steps of S1 are as follows:

[0041] S11: Obtain information on the association between microorganisms and diseases, and collect information on the association between microorganisms and drugs, and drugs and diseases, and delete duplicate entries and non-human disease entries in the collected information;

[0042] S12: Based on the collected information on the association between known microorganisms and diseases, construct the microorganism-disease adjacency matrix A1; based on the collected information on the association between known microorganisms and drugs, and drugs and diseases, construct the microorganism-disease adjacency matrix A2.

[0043] S13: Construct a comprehensive microbial similarity matrix ZM1 based on A1, and a comprehensive microbial similarity matrix ZM2 based on A2; since the construction process of ZM1 is similar to that of ZM2, ZM2 is adopted. v Let v = {1, 2} to represent ZM1 or ZM2. When v = 1, ZM v ZM1 represents ZM; when v = 2, ZM v Indicates ZM2;

[0044] S14: Construct a comprehensive disease similarity matrix ZD1 based on A1, and a comprehensive disease similarity matrix ZD2 based on A2; since the construction process of ZD1 is similar to that of ZD2, ZD2 is used. v Let v = {1, 2} to represent ZD1 or ZD2. When v = 1, ZD v ZD1 represents ZD; when v = 2, ZD v Indicates ZD2;

[0045] S15: Based on matrices a1, ZM1, and ZD1, construct the first microorganism-disease heterogeneous association graph, represented as a matrix. Based on matrices a2, ZM2, and ZD2, a second microorganism-disease heterogeneous association graph is constructed, represented as a matrix. The specific construction process is as follows:

[0046]

[0047] As a preferred embodiment of the microbial-disease association prediction method based on complementary fusion of dual association graphs described in this invention, the specific steps of S12 are as follows:

[0048] S121: Based on the collected known association information between microorganisms and diseases, let nm and nd represent the number of microorganisms and diseases, respectively, to obtain the microorganism-disease adjacency matrix. The specific construction method is as follows: For any given microorganism m i and disease dj If there is a known association between them, then A1(i,j) = 1; otherwise, A1(i,j) = 0.

[0049] S122: Based on the collected information on known associations between microorganisms and drugs, and between drugs and diseases, construct a microorganism-disease adjacency matrix. The specific construction method is as follows: For any given microorganism m i Disease d j If there exists a known m i With a certain drug r k The relationship between them, and r k With d j If there is a known relationship between them, then A2(i,j) = 1; otherwise, A2(i,j) = 0.

[0050] As a preferred embodiment of the microbial-disease association prediction method based on complementary fusion of dual association graphs described in this invention, wherein: ZM in S13 v The construction process is as follows:

[0051] S131: Construct the microbial GIP kernel similarity matrix GM1 based on A1, and construct the microbial GIP kernel similarity matrix GM2 based on A2; since the construction process of GM1 is similar to that of GM2, GM1 is used. v Let v = {1, 2} to represent GM1 or GM2. When v = 1, GM1 is equal to GM2. v GM1 represents GM; when v = 2, GM v GM2 is represented by this matrix; the microbial GIP nuclear similarity matrix GM is constructed using Gaussian nuclear interaction spectra. v The specific calculation formula is as follows:

[0052]

[0053] Where λ1 represents the normalized kernel bandwidth; |||| F Let A denote the Frobenius norm; in the following text, for ease of describing the general calculation process, A will be used. v v = {1, 2} represents A1 or A2, that is, when v = 1, A v Represent A1; when v = 2, A v Represents A2; IP v (m i ) and IP v (m j ) are based on A v Constructed microorganisms m i and microorganisms m j The binary vectors, respectively recording m i and m jInteractions with all diseases;

[0054] S132: Construct the microbial cosine similarity matrix CM1 based on A1, and construct the microbial cosine similarity matrix CM2 based on A2; since the construction process of CM1 is similar to that of CM2, CM2 is used. v Let v = {1, 2} to represent CM1 and CM2. When v = 1, CM v CM1 represents CM; when v = 2, CM v Represent CM2; using m i and m j Constructing a microbial cosine similarity matrix (CM) based on the cosine similarity between them. v The specific calculation formula is as follows:

[0055]

[0056] Among them, A v (i,:) indicates that in A v The association vector between node i and the disease; A v (j,:) indicates that in A v The association vector between node j and the disease;

[0057] S133: Based on GM v and CM v Microbial comprehensive similarity calculations were performed to obtain microbial comprehensive similarity matrices ZM1 and ZM2; m i and m j The formula for calculating the overall similarity between them is as follows:

[0058]

[0059] As a preferred embodiment of the microbial-disease association prediction method based on complementary fusion of dual association graphs described in this invention, wherein: ZD in S14 v The construction process is as follows:

[0060] S141: Construct the disease GIP kernel similarity matrix GD1 based on A1, and construct the disease GIP kernel similarity matrix GD2 based on A2; since the construction process of GD1 is similar to that of GD2, GD2 is used. v Let v = {1, 2} to represent GD1 or GD2. When v = 1, GD v GD1 represents GD; when v = 2, GD v GD2 is represented; a disease GIP kernel similarity matrix GD is constructed using the Gaussian nuclear interaction spectrum (GIP). v The specific calculation formula is as follows:

[0061]

[0062] Where λ2 is the normalized kernel bandwidth; IP v (d i ) and IP v (d j ) are based on A v Constructed disease d i and disease d j The binary vectors represent d respectively. i and d j Interactions with all microorganisms;

[0063] S142: Construct a cosine similarity matrix CD1 based on A1, and a cosine similarity matrix CD2 based on A2; since the construction process of CD1 is similar to that of CD2, CD2 is used. v Let v = {1, 2} to represent CD1 and CD2. When v = 1, CD1 and CD2 are represented by v = {1, 2}. v CD1 represents CD; when v = 2, CD v Represent CD2; using d i and d j Constructing a cosine similarity matrix (CD) between diseases based on their cosine similarity. v The specific calculation formula is as follows:

[0064]

[0065] Among them, A v (:,i) indicates that in A v The association vector between node i and the microorganism; A v (j,:) indicates that in A v The association vector between node j and the microorganism;

[0066] S143: Based on GD v and CD v Perform comprehensive disease similarity calculations to obtain comprehensive disease similarity matrices ZD1 and ZD2; d i and d j The formula for calculating the overall similarity between them is as follows:

[0067]

[0068] As a preferred embodiment of the microbial-disease association prediction method based on complementary fusion of dual association graphs described in this invention, the specific steps of S2 are as follows:

[0069] S21: Based on protein function associations, calculate for any given m i and m j Functional similarity between them, constructing a microbial functional similarity matrix. Based on gene interaction information, the calculation is performed for any given d. i and d j Functional similarity between them, constructing a disease functional similarity matrix

[0070] S22: Perform a topology-enhancing restart random walk operation TERRW on matrices ZM1, ZM2, ZD1, and ZD2 respectively to obtain the first microbial topology attribute matrix. Second microbial topological attribute matrix First Disease Topological Attribute Matrix Second disease topological attribute matrix The TERRW operation is defined as follows:

[0071]

[0072] in, c represents the walk probability distribution of the i-th node in the l-th step; i It is the dynamic adjustment probability calculated by node i based on its local topology information; It is the extended transition probability matrix; ∈ i It is the initial probability vector of node i;

[0073] S23: To preserve more original features, the functional similarity, topological attributes, and adjacency relationships of microorganisms and diseases are concatenated to construct the feature matrix corresponding to B1. The characteristic matrix corresponding to B2 Its Z1 and Z2 are defined as follows:

[0074]

[0075] Where A1 represents the microorganism-disease adjacency matrix A1, and A2 represents the microorganism-disease adjacency matrix A2.

[0076] As a preferred embodiment of the microbial-disease association prediction method based on complementary fusion of dual association graphs described in this invention, in step S22, since the construction process of MM2, DD1, and DD2 is similar to that of MM1, the construction process of MM1 is as follows:

[0077] S221: Set the initial probability vector of node i ∈ i The specific construction formula is as follows:

[0078]

[0079] S222: Based on ZM1, the transition probability matrix M is constructed by normalizing the elements in each row. The specific construction formula is as follows:

[0080]

[0081] Construct the two-hop transition probability matrix M based on M. 2 This is used to capture the two-hop neighbor information of a node, and the specific construction formula is as follows:

[0082] M 2 =M×M;

[0083] S223: By adjusting (M+M) 2 The extended transition probability matrix is ​​obtained by normalization.

[0084] S224: For each node i, calculate the dynamic adjustment probability c based on its local topology. i The specific calculation process is as follows:

[0085]

[0086] Where deg(i) is the degree of node i. Let i represent the set of neighbors of node i;

[0087] S225: Using TERRW's formula, iteratively update the probability distribution of nodes until the probability distribution converges; the specific calculation formula is as follows:

[0088]

[0089] S226: Integrate the probability vectors of all nodes into MM1. The specific construction process is as follows:

[0090] MM1 = [p1, p2, ... p i ,…,p nm ]

[0091] Where, p i It is the probability distribution of node i after TERRW convergence.

[0092] As a preferred embodiment of the microbial-disease association prediction method based on complementary fusion of dual association graphs described in this invention, the specific steps of S3 are as follows:

[0093] S31: Construct a dual-association graph fusion matrix B based on B1 and B2;

[0094] S32: Based on B, apply the dual association graph complementary attention mechanism DAG-CA to calculate the attention weight between each node and its neighboring nodes in the same association graph and the attention weight between its neighboring nodes across association graphs.

[0095] S33: Combine the neighbor node weights obtained from Z1, Z2 and DAG-CA to perform bi-association graph information fusion and construct the node collaborative representation matrix H corresponding to B;

[0096] S34: The model uses the Adam optimizer, minimizing the soft-label KL divergence loss L. KL To train collaborative representation learning; soft-label KL divergence loss L KL It is used to measure the difference between the target distribution and the soft label assignment distribution, and is defined as follows:

[0097]

[0098] Where C is the number of clusters; t eo This is the soft label assignment probability, representing the probability that node e belongs to cluster o; q eo It is the target distribution, obtained by squared normalization of the soft label assignment probability;

[0099] t eo The calculation process is as follows:

[0100]

[0101] Where, μ o The center of cluster o is randomly initialized, and for each cluster center it is initialized as a random node in H; ||*|| represents the Euclidean distance;

[0102] q eo The calculation process is as follows:

[0103]

[0104] As a preferred embodiment of the microbial-disease association prediction method based on complementary fusion of dual association graphs described in this invention, the specific steps of S31 are as follows:

[0105] S311: Construct a cross-association graph connection matrix B cross The formula for calculating the value of any element in the array is as follows:

[0106]

[0107] in, B cross Does node i have a cross-association graph connection with node j?

[0108] S312: Based on B1, B2 and B cross Construct a dual-association graph fusion matrix The specific construction process is as follows:

[0109]

[0110] The DAG-CA execution process in S32 is as follows:

[0111] S321: Transform Z1 into query matrix Q1 and key matrix K1 using linear transformations, and transform Z2 into query matrix Q2 and key matrix K2 using linear transformations; the specific transformation process is as follows:

[0112]

[0113] in, These are the learned weight matrix parameters, all of which have a size of [missing information].

[0114] S322: Based on B, calculate the attention weight α between node i in B1 and its neighbor node u in the same association graph. iu ;α iu The specific calculation formula is as follows:

[0115]

[0116] Where, d k yes and Dimensions, This represents a scaling factor; This represents the query vector corresponding to node i in the Q1 matrix; Let N1(i) represent the key vector of the corresponding neighbor node u in matrix K1; N1(i) represent the set of neighbors of node i in matrix B1.

[0117] S323: Based on B, calculate the attention weight γ between node i and its neighbor node u in the same association graph in B2. iu ;γ iu The specific calculation formula is as follows:

[0118]

[0119] in, This represents the query vector corresponding to node i in the Q2 matrix; Let N2(i) represent the key vector of the corresponding neighbor node u in matrix K2; let N2(i) represent the set of neighbors of node i in matrix B2.

[0120] S324: Based on B, calculate the attention weight β between node i in B1 and its neighbor node n across the association graph. in ; The cross-association graph neighbor nodes of node i are connected through B in B cross B2 can be obtained in the following ways: If Then the node corresponding to the non-zero position in column t of B2 is the cross-association graph neighbor node of i; β in The specific calculation formula is as follows:

[0121]

[0122] Comp-Softmax is an activation function that focuses on irrelevant supplementary information across the association graph. It enhances irrelevant information and weakens relevant information, as shown in the following formula:

[0123] Comp-softmax(R) = softmax(-R)

[0124] Where R represents the strength of the relationship between nodes;

[0125] S325: Based on B, calculate the attention weight δ between node i in B2 and its neighbor node n across the association graph. in ; The cross-association graph neighbor nodes of node i are connected through B in B cross T And obtain it from B1, the specific method of obtaining it is as follows: If Then the node corresponding to the non-zero position in column t of B1 is the cross-association graph neighbor node of i; δ in The specific calculation formula is as follows:

[0126]

[0127] The specific steps of S33 are as follows:

[0128] S331: Based on the nodes in B1, perform bi-association graph information fusion to obtain matrix H1. The information fusion process corresponding to node i is as follows:

[0129]

[0130] in, Let W1 be the node representation vector of node i in B1 after information fusion of the two association graphs; N1(i) and N2(i) represent the neighbor sets of node i in B1 and B2, respectively; |N1(i)| and |N2(i)| represent the number of neighbors of node i in B1 and B2, respectively; W1, W2, and W3 are the learned weight matrices. Let represent the initial feature vector of node i in B1. Let represent the initial feature vector of neighbor node u in B1. Let σ represent the initial feature vector of neighbor node n in B2; σ is the activation function.

[0131] S332: Based on the nodes in B2, perform bi-association graph information fusion to obtain matrix H2. The information fusion process corresponding to node i is as follows:

[0132]

[0133] in, W1 is the node representation vector of node i in B2 after information fusion of the two association graphs; W4, W5, and W6 are the learned weight matrices.

[0134] S333: Construct the node collaborative representation matrix H corresponding to B, and its calculation formula is as follows:

[0135]

[0136] As a preferred embodiment of the microbial-disease association prediction method based on complementary fusion of dual association graphs described in this invention, the specific steps of S4 are as follows:

[0137] S41: Using H, define the collaborative representation vector H for each microorganism. i Collaborative representation vector H with each disease j The correlation between them to MDA ij MDA ij The calculation is as follows:

[0138] MDA ij =-||H i -H j ||

[0139] Among them, H i -H j This indicates the difference between a microorganism and a disease; the smaller the difference, the higher the likelihood that the microorganism-disease pair is associated.

[0140] S42: Utilizing MDA ij Using MLP to determine the predictive association probability of microorganisms-diseases ij The specific methods are as follows:

[0141] ass ij =MLP(MDA) ij ).

[0142] Compared with existing technologies:

[0143] This invention effectively integrates multi-dimensional feature information through a graph-enhanced feature fusion extraction module, improving the accuracy and comprehensiveness of feature representation. Simultaneously, by combining topology-enhanced Restarted Random Walk (TERRW), it enhances the global perception capability of node representation, enabling more accurate capture of complex relationships between nodes. Furthermore, the dual-association graph complementary fusion model, based on the microorganism-disease dual-association graph fusion matrix, combines a dual-association graph complementary attention mechanism and a soft-label KL divergence loss function to effectively address the problem of insufficient information fusion between different association graphs, overcoming the limitation of restricted complementarity between association graphs, thereby significantly improving information fusion effect and prediction accuracy. At the same time, the model can capture potential nonlinear and cross-domain interactions between microorganisms and diseases, improving the model's robustness, stability, and interpretability, making it particularly suitable for processing complex heterogeneous network data. Attached Figure Description

[0144] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0145] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0146] This invention provides a method for predicting the association between microorganisms and diseases based on complementary fusion of dual association graphs. Please refer to [link to relevant documentation]. Figure 1 The specific steps are as follows:

[0147] S1: Obtain the association information between microorganisms and diseases, construct the first microorganism-disease heterogeneous association graph, represented as matrix B1; obtain the association information between microorganisms and drugs, and drugs and diseases, construct the second microorganism-disease heterogeneous association graph, represented as matrix B2;

[0148] The specific steps of S1 are as follows:

[0149] S11: Obtain information on the association between microorganisms and diseases, and collect information on the association between microorganisms and drugs, and drugs and diseases, and delete duplicate entries and non-human disease entries in the collected information;

[0150] S11 includes, but is not limited to, the following embodiments:

[0151] We downloaded known associations between microorganisms and diseases from the public database MDAD. Among the 73 microorganisms and 109 human diseases collected, 502 microorganisms had documented associations with diseases. We also downloaded known associations between microorganisms and drugs, and between drugs and diseases from the DrugBank database. We collected 1470 associations between 73 microorganisms and 686 drugs, and 1121 associations between 233 drugs and 109 corresponding diseases.

[0152] S12: Based on the collected information on the association between known microorganisms and diseases, construct the microorganism-disease adjacency matrix A1; based on the collected information on the association between known microorganisms and drugs, and drugs and diseases, construct the microorganism-disease adjacency matrix A2.

[0153] The specific steps in S12 are as follows:

[0154] S121: Based on the collected known association information between microorganisms and diseases, let nm and nd represent the number of microorganisms and diseases, respectively, to obtain the microorganism-disease adjacency matrix. The specific construction method is as follows: For any given microorganism m i and disease d j If there is a known association between them, then A1(i,j) = 1; otherwise, A1(i,j) = 0.

[0155] S122: Based on the collected information on known associations between microorganisms and drugs, and between drugs and diseases, construct a microorganism-disease adjacency matrix. The specific construction method is as follows: For any given microorganism m i Disease d j If there exists a known m i With a certain drug r k The relationship between them, and r k With d j If there is a known relationship between them, then A2(i,j) = 1; otherwise, A2(i,j) = 0.

[0156] S13: Construct a comprehensive microbial similarity matrix ZM1 based on A1, and a comprehensive microbial similarity matrix ZM2 based on A2; since the construction process of ZM1 is similar to that of ZM2, ZM2 is adopted. v Let v = {1, 2} to represent ZM1 or ZM2. When v = 1, ZM v ZM1 represents ZM; when v = 2, ZM v Indicates ZM2;

[0157] Among them, ZM in S13 v The construction process is as follows:

[0158] S131: Construct the microbial GIP kernel similarity matrix GM1 based on A1, and construct the microbial GIP kernel similarity matrix GM2 based on A2; since the construction process of GM1 is similar to that of GM2, GM1 is used. v Let v = {1, 2} to represent GM1 or GM2. When v = 1, GM1 is equal to GM2. v GM1 represents GM; when v = 2, GM v GM2 is represented by this matrix; the microbial GIP nuclear similarity matrix GM is constructed using Gaussian nuclear interaction spectra.v The specific calculation formula is as follows:

[0159]

[0160] Where λ1 represents the normalized kernel bandwidth; |||| F Let A denote the Frobenius norm; in the following text, for ease of describing the general calculation process, A will be used. v v = {1, 2} represents A1 or A2, that is, when v = 1, A v Represent A1; when v = 2, A v Represents A2; IP v (m i ) and IP v (m j ) are based on A v Constructed microorganisms m i and microorganisms m j The binary vectors, respectively recording m i and m j Interactions with all diseases;

[0161] Among them, Gaussian nuclear interaction spectrum is a publicly available method for representing the similarity between species; GIP kernel similarity matrix can quantify the similarity between different microorganisms and between different diseases through Gaussian kernel function; Frobenius norm is an important concept in matrix analysis, often used to measure the "size" or "length" of a matrix, and it is the square root of the sum of the squares of the matrix elements;

[0162] S132: Construct the microbial cosine similarity matrix CM1 based on A1, and construct the microbial cosine similarity matrix CM2 based on A2; since the construction process of CM1 is similar to that of CM2, CM2 is used. v Let v = {1, 2} to represent CM1 and CM2. When v = 1, CM v CM1 represents CM; when v = 2, CM v Represent CM2; using m i and m j Constructing a microbial cosine similarity matrix (CM) based on the cosine similarity between them. v The specific calculation formula is as follows:

[0163]

[0164] Among them, A v (i,:) represents A v The i-th row, A v (j,:) represents A v The j-th row;

[0165] Cosine similarity is a metric for measuring the similarity between two vectors; it measures their similarity by calculating the cosine of the angle between the two vectors, focusing primarily on the direction of the vectors rather than their magnitude.

[0166] S133: Based on GM v and CM v Microbial comprehensive similarity calculations were performed to obtain microbial comprehensive similarity matrices ZM1 and ZM2; m i and m j The formula for calculating the overall similarity between them is as follows:

[0167]

[0168] Among them, comprehensive similarity is a concept used to measure the degree of similarity between two or more objects. It does not rely on a single similarity to judge, but rather uses a combination of multiple similarity calculation methods to more comprehensively and accurately assess the similarity relationship between objects.

[0169] S14: Construct a comprehensive disease similarity matrix ZD1 based on A1, and a comprehensive disease similarity matrix ZD2 based on A2; since the construction process of ZD1 is similar to that of ZD2, ZD2 is used. v Let v = {1, 2} to represent ZD1 or ZD2. When v = 1, ZD v ZD1 represents ZD; when v = 2, ZD v Indicates ZD2;

[0170] Among them, ZD in S14 v The construction process is as follows:

[0171] S141: Construct the disease GIP kernel similarity matrix GD1 based on A1, and construct the disease GIP kernel similarity matrix GD2 based on A2; since the construction process of GD1 is similar to that of GD2, GD2 is used. v Let v = {1, 2} to represent GD1 or GD2. When v = 1, GD v GD1 represents GD; when v = 2, GD v GD2 is represented; a disease GIP kernel similarity matrix GD is constructed using the Gaussian nuclear interaction spectrum (GIP). v The specific calculation formula is as follows:

[0172]

[0173] Where λ2 is the normalized kernel bandwidth; IP v (d i ) and IP v (d j ) are based on A v Constructed disease d iand disease d j The binary vectors represent d respectively. i and d j Interactions with all microorganisms;

[0174] S142: Construct a cosine similarity matrix CD1 based on A1, and a cosine similarity matrix CD2 based on A2; since the construction process of CD1 is similar to that of CD2, CD2 is used. v Let v = {1, 2} to represent CD1 and CD2. When v = 1, CD1 and CD2 are represented by v = {1, 2}. v CD1 represents CD; when v = 2, CD v Represent CD2; using d i and d j Constructing a cosine similarity matrix (CD) between diseases based on their cosine similarity. v The specific calculation formula is as follows:

[0175]

[0176] Among them, A v (:,i) indicates that in A v The association vector between node i and the microorganism; A v (j,:) indicates that in A v The association vector between node j and the microorganism;

[0177] S143: Based on GD v and CD v Perform comprehensive disease similarity calculations to obtain comprehensive disease similarity matrices ZD1 and ZD2; d i and d j The formula for calculating the overall similarity between them is as follows:

[0178]

[0179] S15: Based on matrices A1, ZM1, and ZD1, construct the first microorganism-disease heterogeneous association graph, represented as a matrix. Based on matrices A2, ZM2, and ZD2, a second microorganism-disease heterogeneous association graph is constructed, represented as a matrix. The specific construction process is as follows:

[0180]

[0181] S2: Apply the graph enhancement feature fusion extraction module GEFFE to obtain the feature matrix Z1 corresponding to B1 and the feature matrix Z2 corresponding to B2.

[0182] The specific steps of S2 are as follows:

[0183] S21: Based on protein function associations, calculate for any given m i and m j Functional similarity between them, constructing a microbial functional similarity matrix. Based on gene interaction information, the calculation is performed for any given d. i and d j Functional similarity between them, constructing a disease functional similarity matrix

[0184] The method for calculating the functional similarity between microorganisms based on protein-protein functional associations is a publicly available method and not the subject of this invention. This method constructs a protein-protein functional association network among microorganisms to calculate the functional similarity (i.e., the microbial functional association index, MFI). First, functional association information of proteins in the microbial genome is extracted, and nodes (gene families) in the network are labeled according to the presence or absence of gene families. Then, the connections (edges) between different types of gene families are counted, and finally, the functional similarity between two microorganisms is calculated using a formula. This method can be used to generate a functional similarity matrix between microorganisms, reflecting the degree of functional association between different microorganisms.

[0185] The method for calculating the functional similarity between diseases based on gene interaction information is a previously disclosed method and is not the subject of this invention. First, the relevant gene set is derived for each disease. Then, the similarity between diseases is measured based on the functional similarity between genes (assessed by log-likelihood score). Specifically, the similarity between disease pairs is obtained by calculating the maximum functional similarity between the relevant gene sets and comprehensively considering the functional associations of all genes, ultimately yielding a functional similarity matrix of diseases.

[0186] S22: Perform a topology-enhanced restarted random walk (TERRW) operation on matrices ZM1, ZM2, ZD1, and ZD2 respectively to obtain the first microbial topological attribute matrix. Second microbial topological attribute matrix First Disease Topological Attribute Matrix Second disease topological attribute matrix The TERRW operation is defined as follows:

[0187]

[0188] in, c represents the walk probability distribution of the i-th node in the l-th step; i It is the dynamic adjustment probability calculated by node i based on its local topology information; It is the extended transition probability matrix; ∈i It is the initial probability vector of node i;

[0189] In S22, since the construction process of MM2, DD1, and DD2 is similar to that of MM1, the construction process of MM1 is as follows:

[0190] S221: Set the initial probability vector of node i ∈ i The specific construction formula is as follows:

[0191]

[0192] S222: Based on ZM1, the transition probability matrix M is constructed by normalizing the elements in each row. The specific construction formula is as follows:

[0193]

[0194] Construct the two-hop transition probability matrix M based on M. 2 This is used to capture the two-hop neighbor information of a node, and the specific construction formula is as follows:

[0195] M 2 =M×M;

[0196] S223: By adjusting (M+M) 2 The extended transition probability matrix is ​​obtained by normalization.

[0197] S224: For each node i, calculate the dynamic adjustment probability c based on its local topology. i The specific calculation process is as follows:

[0198]

[0199] Where deg(i) is the degree of node i. Let i represent the set of neighbors of node i;

[0200] S225: Using TERRW's formula, iteratively update the probability distribution of nodes until the probability distribution converges; the specific calculation formula is as follows:

[0201]

[0202] S226: Integrate the probability vectors of all nodes into MM1. The specific construction process is as follows:

[0203] MM1 = [p1, p2, ... p i ,…,p nm ]

[0204] Where, p i It is the probability distribution of node i after TERRW convergence;

[0205] S23: To preserve more original features, the functional similarity, topological attributes, and adjacency relationships of microorganisms and diseases are concatenated to construct the feature matrix corresponding to B1. The characteristic matrix corresponding to B2 Its Z1 and Z2 are defined as follows:

[0206]

[0207] Where A1 represents the microorganism-disease adjacency matrix A1, and A2 represents the microorganism-disease adjacency matrix A2;

[0208] S3: Using B1 and B2, construct the dual-association graph fusion matrix B. Simultaneously, based on B, Z1, and Z2, apply the dual-association graph complementary fusion model DAG-CFM to update node information, obtaining the node collaborative representation matrix H corresponding to B. The specific steps are as follows:

[0209] S31: Construct a dual-association graph fusion matrix B based on B1 and B2;

[0210] The specific steps in S31 are as follows:

[0211] S311: Construct a cross-association graph connection matrix B cross The formula for calculating the value of any element in the array is as follows:

[0212]

[0213] in, B cross Does node i have a cross-association graph connection with node j? S312: Based on B1, B2, and B cross Construct a dual-association graph fusion matrix The specific construction process is as follows:

[0214]

[0215] S32: Based on B, apply the dual association graph complementary attention mechanism DAG-CA to calculate the attention weight between each node and its neighboring nodes in the same association graph and the attention weight between its neighboring nodes across association graphs.

[0216] The DAG-CA execution process in S32 is as follows:

[0217] S321: Transform Z1 into query matrix Q1 and key matrix K1 using linear transformations, and transform Z2 into query matrix Q2 and key matrix K2 using linear transformations; the specific transformation process is as follows:

[0218]

[0219] in, These are the learned weight matrix parameters, all of which have a size of [missing information].

[0220] S322: Based on B, calculate the attention weight α between node i in B1 and its neighbor node u in the same association graph. iu ;α iu The specific calculation formula is as follows:

[0221]

[0222] Where, d k yes and Dimensions, This represents a scaling factor; This represents the query vector corresponding to node i in the Q1 matrix; Let N1(i) represent the key vector of the corresponding neighbor node u in matrix K1; N1(i) represent the set of neighbors of node i in matrix B1.

[0223] S323: Based on B, calculate the attention weight γ between node i and its neighbor node u in the same association graph in B2. iu ;γ iu The specific calculation formula is as follows:

[0224]

[0225] in, This represents the query vector corresponding to node i in the Q2 matrix; Let N2(i) represent the key vector of the corresponding neighbor node u in matrix K2; let N2(i) represent the set of neighbors of node i in matrix B2.

[0226] S324: Based on B, calculate the attention weight β between node i in B1 and its neighbor node n across the association graph. in ; The cross-association graph neighbor nodes of node i are connected through B in B cross Obtained from B2, the specific methods are as follows: If Then the node corresponding to the non-zero position in column t of B2 is the cross-association graph neighbor node of i; β in The specific calculation formula is as follows:

[0227]

[0228] Comp-Softmax is an activation function that focuses on irrelevant supplementary information across the association graph. It enhances irrelevant information and weakens relevant information, as shown in the following formula:

[0229] Comp-softmax(R) = softmax(-R)

[0230] Where R represents the strength of the relationship between nodes;

[0231] S325: Based on B, calculate the attention weight δ between node i in B2 and its neighbor node n across the association graph. in ; The cross-association graph neighbor nodes of node i are connected through B in B cross T And obtain it from B1, the specific method of obtaining it is as follows: If Then the node corresponding to the non-zero position in column t of B1 is the cross-association graph neighbor node of i; δ in The specific calculation formula is as follows:

[0232]

[0233] S33: Combine the neighbor node weights obtained from Z1, Z2 and DAG-CA to perform bi-association graph information fusion and construct the node collaborative representation matrix H corresponding to B;

[0234] The specific steps of S33 are as follows:

[0235] S331: Based on the nodes in B1, perform bi-association graph information fusion to obtain matrix H1. The information fusion process corresponding to node i is as follows:

[0236]

[0237] in, Let be the node representation vector of node i in B1 after information fusion of the two association graphs; N1(i) and N2(i) represent the neighbor sets of node i in B1 and B2, respectively; |N1(i)| and |N2(i)| represent the number of neighbors of node i in B1 and N2, respectively; W1, W2, and W3 are the learned weight matrices. Let represent the initial feature vector of node i in B1. Let represent the initial feature vector of neighbor node u in B1. This represents the initial feature vector of neighbor node n in N2; σ is the activation function;

[0238] S332: Based on the nodes in B2, perform bi-association graph information fusion to obtain matrix H2. The information fusion process corresponding to node i is as follows:

[0239]

[0240] in, W1 is the node representation vector of node i in B2 after information fusion of the two association graphs; W4, W5, and W6 are the learned weight matrices.

[0241] S333: Construct the node collaborative representation matrix H corresponding to B, and its calculation formula is as follows:

[0242]

[0243] S34: The model uses the Adam optimizer, minimizing the soft-label KL divergence loss L. KL To train collaborative representation learning; soft-label KL divergence loss L KL It is used to measure the difference between the target distribution and the soft label assignment distribution, and is defined as follows:

[0244]

[0245] Where C is the number of clusters; t eo This is the soft label assignment probability, representing the probability that node e belongs to cluster o; q eo It is the target distribution, obtained by squared normalization of the soft label assignment probability;

[0246] t eo The calculation process is as follows:

[0247]

[0248] Where, μ o The center of cluster o is randomly initialized, and for each cluster center it is initialized as a random node in H; ||*|| represents the Euclidean distance;

[0249] q eo The calculation process is as follows:

[0250]

[0251] Adam is a publicly available method that combines the advantages of momentum optimization and RMSProp optimization, and can dynamically adjust the learning rate of each parameter.

[0252] Euclidean distance is a publicly available and commonly used distance metric used to measure the straight-line distance between two points in Euclidean space; it is a metric for the shortest path distance between two points as defined in geometry.

[0253] S4: Using H, predict the association between microorganisms and diseases through MLP;

[0254] The specific steps of S4 are as follows:

[0255] S41: Using H, define the collaborative representation vector H for each microorganism. iCollaborative representation vector H with each disease j The correlation between them to MDA ij MDA ij The calculation is as follows:

[0256] MDA ij =-||H i -H j ||

[0257] Among them, H i -H j This indicates the difference between a microorganism and a disease; the smaller the difference, the higher the likelihood that the microorganism-disease pair is associated.

[0258] S42: Utilizing MDA ij Using MLP to determine the predictive association probability of microorganisms-diseases ij The specific methods are as follows:

[0259] ass ij =MLP(MDA) ij );

[0260] MLP stands for Multilayer Perceptron, a publicly available simple neural network model consisting of two fully connected neural network layers used to process data and learn the complex relationship between input and output.

[0261] Based on the above, this invention significantly enhances the diversity and depth of feature representation by integrating multiple information such as functional similarity, topological attributes, and adjacency relationships between microorganisms and diseases. The module first calculates the functional similarity between microorganisms and diseases, and then extracts the topological attributes of nodes using the TERRW (Restarted Random Walk) method with topological enhancement, thereby better capturing the complex global and local relationships in the graph. Finally, the module concatenates these multidimensional features, preserving the original feature information while enhancing its expressive power, enabling the model to more comprehensively reflect the complex association between microorganisms and diseases, and improving the performance and accuracy of subsequent prediction tasks. Furthermore, this invention aims to integrate information from... The model leverages information from two association graphs to enhance node representation capabilities. Specifically, based on the dual association graph fusion matrix, a dual association graph complementary attention mechanism (DAG-CA) is designed. This mechanism assigns different weights to neighboring nodes within the same association graph and across different association graphs, ensuring full utilization of the complementarity between association graphs during information fusion. Furthermore, during training, a soft-label KL divergence loss optimization strategy is employed to ensure the accuracy of soft label assignment, thereby enhancing the collaborative representation capabilities of nodes. This model can fully integrate information from two association graphs, improving its stability, efficiency, and interpretability. It is particularly suitable for processing complex heterogeneous network data such as microorganism-disease networks, demonstrating significant performance advantages.

[0262] Although the present invention has been described above with reference to embodiments, various modifications can be made and components can be replaced with equivalents without departing from the scope of the invention. In particular, as long as there is no structural conflict, the features in the disclosed embodiments can be combined with each other in any manner. The lack of an exhaustive description of these combinations in this specification is merely for the sake of brevity and resource conservation. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for predicting the association between microorganisms and diseases based on complementary fusion of dual association graphs, characterized in that, The specific steps are as follows: S1: Obtain the association information between microorganisms and diseases, construct the first microorganism-disease heterogeneous association graph, represented as matrix B1; obtain the association information between microorganisms and drugs, and drugs and diseases, construct the second microorganism-disease heterogeneous association graph, represented as matrix B2; S2: Apply the graph-enhanced feature fusion extraction module GEFFE to obtain the feature matrix Z1 corresponding to B1 and the feature matrix Z2 corresponding to B2. The specific steps are as follows: S21: Based on protein function associations, calculate for any given m i and m j Functional similarity between them, constructing a microbial functional similarity matrix. Based on gene interaction information, the calculation is performed for any given d. i and d j Functional similarity between them, constructing a disease functional similarity matrix Where nm and nd represent the number of microorganisms and diseases, respectively; S22: Perform a topology-enhancing restart random walk operation TERRW on matrices ZN1, ZM2, ZD1, and ZD2 respectively to obtain the first microbial topology attribute matrix. Second microbial topological attribute matrix First Disease Topological Attribute Matrix Second disease topological attribute matrix The TERRW operation is defined as follows: in, c represents the walk probability distribution of the i-th node in the l-th step; i It is the dynamic adjustment probability calculated by node i based on its local topology information; It is the extended transition probability matrix; ∈ i It is the initial probability vector of node i; S23: To preserve more original features, the functional similarity, topological attributes, and adjacency relationships between microorganisms and diseases are concatenated to construct the feature matrix corresponding to B1. The characteristic matrix corresponding to B2 Its Z1 and Z2 are defined as follows: Where A1 represents the microorganism-disease adjacency matrix A1, and A2 represents the microorganism-disease adjacency matrix A2; S3: Using B1 and B2, construct the dual-association graph fusion matrix B. Simultaneously, based on B, Z1, and Z2, apply the dual-association graph complementary fusion model DAG-CFM to update node information, obtaining the node collaborative representation matrix H corresponding to B. The specific steps are as follows: S31: Construct a dual-association graph fusion matrix B based on B1 and B2; S32: Based on B, apply the dual association graph complementary attention mechanism DAG-CA to calculate the attention weight between each node and its neighboring nodes in the same association graph and the attention weight between its neighboring nodes across association graphs. S33: Combine the neighbor node weights obtained from Z1, Z2 and DAG-CA to perform bi-association graph information fusion and construct the node collaborative representation matrix H corresponding to B; S34: The model uses the Adam optimizer, minimizing the soft-label KL divergence loss L. KL To train collaborative representation learning; soft-label KL divergence loss L KL It is used to measure the difference between the target distribution and the soft label assignment distribution, and is defined as follows: Where C is the number of clusters; t eo This is the soft label assignment probability, representing the probability that node e belongs to cluster o; q eo It is the target distribution, obtained by squared normalization of the soft label assignment probability; t eo The calculation process is as follows: Where, μ o The center of cluster o is randomly initialized, and for each cluster center it is initialized as a random node in H; ||*|| represents the Euclidean distance; q eo The calculation process is as follows: S4: Using H, predict the association between microorganisms and diseases through MLP. The specific steps are as follows: S41: Using H, define the collaborative representation vector H for each microorganism. i Collaborative representation vector H with each disease j The correlation between them to MDA ij MDA ij The calculation is as follows: MDA ij =-||H i -H j || Among them, H i -H j This indicates the difference between a microorganism and a disease; the smaller the difference, the higher the likelihood that the microorganism-disease pair is associated. S42: Utilizing MDA ij Using MLP to determine the predictive association probability of microorganisms-diseases ij The specific methods are as follows: ass ij =MLP(MDA ij )。 2. The method for predicting the association between microorganisms and diseases based on complementary fusion of dual association graphs according to claim 1, characterized in that, The specific steps of S1 are as follows: S11: Obtain information on the association between microorganisms and diseases, and collect information on the association between microorganisms and drugs, and drugs and diseases, and delete duplicate entries and non-human disease entries in the collected information; S12: Based on the collected information on the association between known microorganisms and diseases, construct the microorganism-disease adjacency matrix A1; based on the collected information on the association between known microorganisms and drugs, and drugs and diseases, construct the microorganism-disease adjacency matrix A2. S13: Construct a comprehensive microbial similarity matrix ZM1 based on A1, and a comprehensive microbial similarity matrix ZM2 based on A2; since the construction process of ZM1 is similar to that of ZM2, ZM2 is adopted. v Let v = {1, 2} to represent ZM1 or ZM2. When v = 1, ZM v ZM1 represents ZM; when v = 2, ZM v Indicates ZM2; S14: Construct a comprehensive disease similarity matrix ZD1 based on A1, and a comprehensive disease similarity matrix ZD2 based on A2; since the construction process of ZD1 is similar to that of ZD2, ZD2 is used. v Let v = {1, 2} to represent ZD1 or ZD2. When v = 1, ZD v ZD1 represents ZD; when v = 2, ZD v Indicates ZD2; S15: Based on matrices A1, ZM1, and ZD1, construct the first microorganism-disease heterogeneous association graph, represented as a matrix. Based on matrices A2, ZM2, and ZD2, a second microorganism-disease heterogeneous association graph is constructed, represented as a matrix. The specific construction process is as follows:

3. The method for predicting the association between microorganisms and diseases based on complementary fusion of dual association graphs according to claim 2, characterized in that, The specific steps of S12 are as follows: S121: Based on the collected known association information between microorganisms and diseases, let nm and nd represent the number of microorganisms and diseases, respectively, to obtain the microorganism-disease adjacency matrix. The specific construction method is as follows: For any given microorganism m i and disease d j If there is a known association between them, then A1(i,j) = 1; otherwise, A1(i,j) = 0. S122: Based on the collected information on known microorganisms and drugs, as well as the association between drugs and diseases, construct a microorganism-disease adjacency matrix. The specific construction method is as follows: For any given microorganism m i Disease d j If there exists a known m i With a certain drug r k The relationship between them, and r k With d j If there is a known relationship between them, then A2(i,j) = 1; otherwise, A2(i,j) = 0.

4. The method for predicting the association between microorganisms and diseases based on complementary fusion of dual association graphs according to claim 2, characterized in that, ZM in S13 v The construction process is as follows: S131: Construct the microbial GIP kernel similarity matrix GM1 based on A1, and construct the microbial GIP kernel similarity matrix GM2 based on A2; since the construction process of GM1 is similar to that of GM2, GM1 is used. v Let v = {1, 2} to represent GM1 or GM2. When v = 1, GM1 is equal to GM2. v GM1 represents GM; when v = 2, GM v GM2 is represented by this matrix; the microbial GIP nuclear similarity matrix GM is constructed using Gaussian nuclear interaction spectra. v The specific calculation formula is as follows: Where λ1 represents the normalized kernel bandwidth; |||| F Let A denote the Frobenius norm; in the following text, for ease of describing the general calculation process, A will be used. v v = {1, 2} represents A1 or A2, that is, when v = 1, A v Represent A1; when v = 2, A v Represents A2; IP v (m i ) and IP v (m j ) are based on A v Constructed microorganisms m i and microorganisms m j The binary vectors, respectively recording m i and m j Interactions with all diseases; S132: Construct the microbial cosine similarity matrix CM1 based on A1, and construct the microbial cosine similarity matrix CM2 based on A2; since the construction process of CM1 is similar to that of CM2, CM2 is used. v Let v = {1, 2} to represent CM1 and CM2. When v = 1, CM v CM1 represents CM; when v = 2, CM v Represent CM2; using m i and m j Constructing a microbial cosine similarity matrix (CM) based on the cosine similarity between them. v The specific calculation formula is as follows: Among them, A v (i,:) indicates that in A v The association vector between node i and the disease; A v (j,:) indicates that in A v The association vector between node j and the disease; S133: Based on GM v and CM v Microbial comprehensive similarity calculations were performed to obtain microbial comprehensive similarity matrices ZM1 and ZM2; m i and m j The formula for calculating the overall similarity between them is as follows:

5. The method for predicting the association between microorganisms and diseases based on complementary fusion of dual association graphs according to claim 2, characterized in that, ZD in S14 v The construction process is as follows: S141: Construct the disease GIP kernel similarity matrix GD1 based on A1, and construct the disease GIP kernel similarity matrix GD2 based on A2; since the construction process of GD1 is similar to that of GD2, GD2 is used. v Let v = {1, 2} to represent GD1 or GD2. When v = 1, GD v GD1 represents GD; when v = 2, GD v GD2 is represented; a disease GIP kernel similarity matrix GD is constructed using the Gaussian nuclear interaction spectrum (GIP). v The specific calculation formula is as follows: Where λ2 is the normalized kernel bandwidth; IP v (d i ) and IP v (d j ) are based on A v Constructed disease d i and disease d j The binary vectors represent d respectively. i and d j Interactions with all microorganisms; S142: Construct a cosine similarity matrix CD1 based on A1, and a cosine similarity matrix CD2 based on A2; since the construction process of CD1 is similar to that of CD2, CD2 is used. v Let v = {1, 2} to represent CD1 and CD2. When v = 1, CD1 and CD2 are represented by v = {1, 2}. v CD1 represents CD; when v = 2, CD v Represent CD2; using d i and d j Constructing a cosine similarity matrix (CD) between diseases based on their cosine similarity. v The specific calculation formula is as follows: Among them, A v (:,i) indicates that in A v The association vector between node i and the microorganism; A v (j,:) indicates that in A v The association vector between node j and microorganisms; S143: Based on GD v and CD v Perform comprehensive disease similarity calculations to obtain comprehensive disease similarity matrices ZD1 and ZD2; d i and d j The formula for calculating the overall similarity between them is as follows:

6. The method for predicting the association between microorganisms and diseases based on complementary fusion of dual association graphs according to claim 1, characterized in that, Since the construction processes of MM2, DD1, and DD2 in S22 are similar to those of MM1, the construction process of MM1 is as follows: S221: Set the initial probability vector of node i ∈ i The specific construction formula is as follows: S222: Based on ZM1, the transition probability matrix M is constructed by normalizing the elements in each row. The specific construction formula is as follows: Construct the two-hop transition probability matrix M based on M. 2 This is used to capture the two-hop neighbor information of a node, and the specific construction formula is as follows: M 2 =M×M; S223: By adjusting (M+M) 2 The extended transition probability matrix is ​​obtained by normalization. S224: For each node i, calculate the dynamic adjustment probability c based on its local topology. i The specific calculation process is as follows: Where deg(i) is the degree of node i. Let i represent the set of neighbors of node i; S225: Using TERRW's formula, iteratively update the probability distribution of nodes until the probability distribution converges; the specific calculation formula is as follows: S226: Integrate the probability vectors of all nodes into MM1. The specific construction process is as follows: MM1=[p1,p2,…p i ,…,p nm ] Where, p i It is the probability distribution of node i after TERRW convergence.

7. The method for predicting the association between microorganisms and diseases based on complementary fusion of dual association graphs according to claim 1, characterized in that, The specific steps of S31 are as follows: S311: Construct a cross-association graph connection matrix B cross The formula for calculating the value of any element in the array is as follows: in, B cross Does node i have a cross-association graph connection with node j? S312: Based on B1, B2 and B cross Construct a dual-association graph fusion matrix The specific construction process is as follows: The DAG-CA execution process in S32 is as follows: S321: Transform Z1 into query matrix Q1 and key matrix K1 using linear transformations, and transform Z2 into query matrix Q2 and key matrix K2 using linear transformations; the specific transformation process is as follows: in, These are the learned weight matrix parameters, all of which have a size of [missing information]. S322: Based on B, calculate the attention weight α between node i in B1 and its neighbor node u in the same association graph. iu ;α iu The specific calculation formula is as follows: Where, d k yes and Dimensions, This represents a scaling factor; This represents the query vector corresponding to node i in the Q1 matrix; Let N1(i) represent the key vector of the corresponding neighbor node u in matrix K1; N1(i) represent the set of neighbors of node i in matrix B1. S323: Based on B, calculate the attention weight γ between node i and its neighbor node u in the same association graph in B2. iu ;γ iu The specific calculation formula is as follows: in, This represents the query vector corresponding to node i in the Q2 matrix; Let N2(i) represent the key vector of the corresponding neighbor node u in matrix K2; let N2(i) represent the set of neighbors of node i in matrix B2. S324: Based on B, calculate the attention weight β between node i in B1 and its neighbor node n across the association graph. in ; The cross-association graph neighbor nodes of node i are connected through B in B cross Obtained from B2, the specific methods are as follows: If Then the node corresponding to the non-zero position in column t of B2 is the cross-association graph neighbor node of i; β in The specific calculation formula is as follows: Comp-Softmax is an activation function that focuses on irrelevant supplementary information across the association graph. It enhances irrelevant information and weakens relevant information, as shown in the following formula: Comp-softmax(R) = softmax(-R) Where R represents the strength of the relationship between nodes; S325: Based on B, calculate the attention weight δ between node i in B2 and its neighbor node n across the association graph. in ; The cross-association graph neighbor nodes of node i are connected through B in B cross T And obtain it from B1, the specific method of obtaining it is as follows: If Then the node corresponding to the non-zero position in column t of B1 is the cross-association graph neighbor node of i; δ in The specific calculation formula is as follows: The specific steps of S33 are as follows: S331: Based on the nodes in B1, perform bi-association graph information fusion to obtain matrix H1. The information fusion process corresponding to node i is as follows: in, Let W1 be the node representation vector of node i in B1 after information fusion of the two association graphs; N1(i) and N2(i) represent the neighbor sets of node i in B1 and B2, respectively; |N1(i)| and |N2(i)| represent the number of neighbors of node i in B1 and B2, respectively; W1, W2, and W3 are the learned weight matrices. Let represent the initial feature vector of node i in B1. Let represent the initial feature vector of neighbor node u in B1. Let σ represent the initial feature vector of neighbor node n in B2; σ is the activation function. S332: Based on the nodes in B2, perform bi-association graph information fusion to obtain matrix H2. The information fusion process corresponding to node i is as follows: in, W1 is the node representation vector of node i in B2 after information fusion of the two association graphs; W4, W5, and W6 are the learned weight matrices. S333: Construct the node collaborative representation matrix H corresponding to B, and its calculation formula is as follows:

Images